Core Concepts

Core Concepts#

This guide introduces the fundamental concepts behind CoreRec and how recommendation systems work.

What is CoreRec?#

CoreRec is a comprehensive framework for building recommendation systems that combines:

  • Deep Learning Models: Neural network-based approaches for capturing complex patterns

  • Collaborative Filtering: User-item interaction-based recommendations

  • Content-Based Filtering: Feature-based recommendations

  • Hybrid Methods: Combining multiple approaches for better results

Core Concepts#

Recommendation Systems#

A recommendation system suggests items to users based on:

  • User preferences: Past interactions, ratings, behavior

  • Item features: Content, metadata, attributes

  • Context: Time, location, device, etc.

Model Types#

1. Collaborative Filtering (CF)#

Predicts user preferences by finding similar users or items.

User-Based CF: “Users similar to you also liked…” Item-Based CF: “Users who liked this also liked…”

2. Content-Based Filtering#

Recommends items similar to what the user has liked before, based on item features.

3. Deep Learning Models#

Neural networks that learn complex patterns from user-item interactions.

4. Hybrid Methods#

Combine multiple approaches to leverage strengths of each.

Key Components#

Base Recommender#

All production models inherit from BaseRecommender, providing:

  • Unified API (fit, predict, recommend)

  • Model persistence (save, load)

  • Consistent error handling

Engines#

CoreRec organizes models under corerec.engines:

  • Top-level engines (corerec.engines.dcn, deepfm, sasrec, …): 14 production models

  • corerec.engines.collaborative: SAR, NCF, FAST, LightGCN, …

  • corerec.engines.content_based: TF-IDF and content methods

  • corerec.sandbox: ~50 experimental research models (not production-tested)

Model Tiers#

CoreRec models are organized into two tiers:

  • Production Models (14 models): Fully tested, CI-enforced, and implement the complete BaseRecommender interface. These are recommended for production deployments. Examples: DCN, DeepFM, SASRec, NCF, LightGCN, TwoTower, FAST.

  • Sandbox Models (~50 models): Experimental implementations for research and learning. These cover a wide range of published algorithms but are not production-tested. Always validate thoroughly before using in production.

See the Model Documentation for the full breakdown.

Workflow#

  1. Data Preparation: Load and preprocess your dataset

  2. Model Selection: Choose appropriate model for your use case

  3. Training: Fit the model on your data

  4. Evaluation: Measure performance metrics

  5. Deployment: Use model for recommendations

Next Steps#

Contents